x86: cpa: move clflush_cache_range()
[linux-2.6/x86.git] / fs / xfs / xfs_attr_leaf.c
blob81f45dae1c57f668c09786b405b2b805862bb556
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_bmap.h"
42 #include "xfs_attr.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_error.h"
47 * xfs_attr_leaf.c
49 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
57 * Routines used for growing the Btree.
59 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
60 xfs_dabuf_t **bpp);
61 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
62 int freemap_index);
63 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
64 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
65 xfs_da_state_blk_t *blk1,
66 xfs_da_state_blk_t *blk2);
67 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
68 xfs_da_state_blk_t *leaf_blk_1,
69 xfs_da_state_blk_t *leaf_blk_2,
70 int *number_entries_in_blk1,
71 int *number_usedbytes_in_blk1);
74 * Routines used for shrinking the Btree.
76 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77 xfs_dabuf_t *bp, int level);
78 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
79 xfs_dabuf_t *bp);
80 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
81 xfs_dablk_t blkno, int blkcnt);
84 * Utility routines.
86 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
87 int src_start,
88 xfs_attr_leafblock_t *dst_leaf,
89 int dst_start, int move_count,
90 xfs_mount_t *mp);
91 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
93 /*========================================================================
94 * Namespace helper routines
95 *========================================================================*/
97 STATIC_INLINE attrnames_t *
98 xfs_attr_flags_namesp(int flags)
100 return ((flags & XFS_ATTR_SECURE) ? &attr_secure:
101 ((flags & XFS_ATTR_ROOT) ? &attr_trusted : &attr_user));
105 * If namespace bits don't match return 0.
106 * If all match then return 1.
108 STATIC_INLINE int
109 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
111 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
115 * If namespace bits don't match and we don't have an override for it
116 * then return 0.
117 * If all match or are overridable then return 1.
119 STATIC_INLINE int
120 xfs_attr_namesp_match_overrides(int arg_flags, int ondisk_flags)
122 if (((arg_flags & ATTR_SECURE) == 0) !=
123 ((ondisk_flags & XFS_ATTR_SECURE) == 0) &&
124 !(arg_flags & ATTR_KERNORMALS))
125 return 0;
126 if (((arg_flags & ATTR_ROOT) == 0) !=
127 ((ondisk_flags & XFS_ATTR_ROOT) == 0) &&
128 !(arg_flags & ATTR_KERNROOTLS))
129 return 0;
130 return 1;
134 /*========================================================================
135 * External routines when attribute fork size < XFS_LITINO(mp).
136 *========================================================================*/
139 * Query whether the requested number of additional bytes of extended
140 * attribute space will be able to fit inline.
141 * Returns zero if not, else the di_forkoff fork offset to be used in the
142 * literal area for attribute data once the new bytes have been added.
144 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
145 * special case for dev/uuid inodes, they have fixed size data forks.
148 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
150 int offset;
151 int minforkoff; /* lower limit on valid forkoff locations */
152 int maxforkoff; /* upper limit on valid forkoff locations */
153 int dsize;
154 xfs_mount_t *mp = dp->i_mount;
156 offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
158 switch (dp->i_d.di_format) {
159 case XFS_DINODE_FMT_DEV:
160 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
161 return (offset >= minforkoff) ? minforkoff : 0;
162 case XFS_DINODE_FMT_UUID:
163 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
164 return (offset >= minforkoff) ? minforkoff : 0;
167 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) {
168 if (bytes <= XFS_IFORK_ASIZE(dp))
169 return mp->m_attroffset >> 3;
170 return 0;
173 dsize = dp->i_df.if_bytes;
175 switch (dp->i_d.di_format) {
176 case XFS_DINODE_FMT_EXTENTS:
178 * If there is no attr fork and the data fork is extents,
179 * determine if creating the default attr fork will result
180 * in the extents form migrating to btree. If so, the
181 * minimum offset only needs to be the space required for
182 * the btree root.
184 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes > mp->m_attroffset)
185 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
186 break;
188 case XFS_DINODE_FMT_BTREE:
190 * If have data btree then keep forkoff if we have one,
191 * otherwise we are adding a new attr, so then we set
192 * minforkoff to where the btree root can finish so we have
193 * plenty of room for attrs
195 if (dp->i_d.di_forkoff) {
196 if (offset < dp->i_d.di_forkoff)
197 return 0;
198 else
199 return dp->i_d.di_forkoff;
200 } else
201 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
202 break;
206 * A data fork btree root must have space for at least
207 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
209 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
210 minforkoff = roundup(minforkoff, 8) >> 3;
212 /* attr fork btree root can have at least this many key/ptr pairs */
213 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
214 maxforkoff = maxforkoff >> 3; /* rounded down */
216 if (offset >= minforkoff && offset < maxforkoff)
217 return offset;
218 if (offset >= maxforkoff)
219 return maxforkoff;
220 return 0;
224 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
226 STATIC void
227 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
229 unsigned long s;
231 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
232 !(XFS_SB_VERSION_HASATTR2(&mp->m_sb))) {
233 s = XFS_SB_LOCK(mp);
234 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
235 XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
236 XFS_SB_UNLOCK(mp, s);
237 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
238 } else
239 XFS_SB_UNLOCK(mp, s);
244 * Create the initial contents of a shortform attribute list.
246 void
247 xfs_attr_shortform_create(xfs_da_args_t *args)
249 xfs_attr_sf_hdr_t *hdr;
250 xfs_inode_t *dp;
251 xfs_ifork_t *ifp;
253 dp = args->dp;
254 ASSERT(dp != NULL);
255 ifp = dp->i_afp;
256 ASSERT(ifp != NULL);
257 ASSERT(ifp->if_bytes == 0);
258 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
259 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
260 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
261 ifp->if_flags |= XFS_IFINLINE;
262 } else {
263 ASSERT(ifp->if_flags & XFS_IFINLINE);
265 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
266 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
267 hdr->count = 0;
268 hdr->totsize = cpu_to_be16(sizeof(*hdr));
269 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
273 * Add a name/value pair to the shortform attribute list.
274 * Overflow from the inode has already been checked for.
276 void
277 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
279 xfs_attr_shortform_t *sf;
280 xfs_attr_sf_entry_t *sfe;
281 int i, offset, size;
282 xfs_mount_t *mp;
283 xfs_inode_t *dp;
284 xfs_ifork_t *ifp;
286 dp = args->dp;
287 mp = dp->i_mount;
288 dp->i_d.di_forkoff = forkoff;
289 dp->i_df.if_ext_max =
290 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
291 dp->i_afp->if_ext_max =
292 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
294 ifp = dp->i_afp;
295 ASSERT(ifp->if_flags & XFS_IFINLINE);
296 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
297 sfe = &sf->list[0];
298 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
299 #ifdef DEBUG
300 if (sfe->namelen != args->namelen)
301 continue;
302 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
303 continue;
304 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
305 continue;
306 ASSERT(0);
307 #endif
310 offset = (char *)sfe - (char *)sf;
311 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
312 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
313 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
314 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
316 sfe->namelen = args->namelen;
317 sfe->valuelen = args->valuelen;
318 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
319 memcpy(sfe->nameval, args->name, args->namelen);
320 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
321 sf->hdr.count++;
322 be16_add(&sf->hdr.totsize, size);
323 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
325 xfs_sbversion_add_attr2(mp, args->trans);
329 * Remove an attribute from the shortform attribute list structure.
332 xfs_attr_shortform_remove(xfs_da_args_t *args)
334 xfs_attr_shortform_t *sf;
335 xfs_attr_sf_entry_t *sfe;
336 int base, size=0, end, totsize, i;
337 xfs_mount_t *mp;
338 xfs_inode_t *dp;
340 dp = args->dp;
341 mp = dp->i_mount;
342 base = sizeof(xfs_attr_sf_hdr_t);
343 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
344 sfe = &sf->list[0];
345 end = sf->hdr.count;
346 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
347 base += size, i++) {
348 size = XFS_ATTR_SF_ENTSIZE(sfe);
349 if (sfe->namelen != args->namelen)
350 continue;
351 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
352 continue;
353 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
354 continue;
355 break;
357 if (i == end)
358 return(XFS_ERROR(ENOATTR));
361 * Fix up the attribute fork data, covering the hole
363 end = base + size;
364 totsize = be16_to_cpu(sf->hdr.totsize);
365 if (end != totsize)
366 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
367 sf->hdr.count--;
368 be16_add(&sf->hdr.totsize, -size);
371 * Fix up the start offset of the attribute fork
373 totsize -= size;
374 if (totsize == sizeof(xfs_attr_sf_hdr_t) && !args->addname &&
375 (mp->m_flags & XFS_MOUNT_ATTR2) &&
376 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE)) {
378 * Last attribute now removed, revert to original
379 * inode format making all literal area available
380 * to the data fork once more.
382 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
383 dp->i_d.di_forkoff = 0;
384 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
385 ASSERT(dp->i_d.di_anextents == 0);
386 ASSERT(dp->i_afp == NULL);
387 dp->i_df.if_ext_max =
388 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
389 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
390 } else {
391 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
392 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
393 ASSERT(dp->i_d.di_forkoff);
394 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) || args->addname ||
395 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
396 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
397 dp->i_afp->if_ext_max =
398 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
399 dp->i_df.if_ext_max =
400 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
401 xfs_trans_log_inode(args->trans, dp,
402 XFS_ILOG_CORE | XFS_ILOG_ADATA);
405 xfs_sbversion_add_attr2(mp, args->trans);
407 return(0);
411 * Look up a name in a shortform attribute list structure.
413 /*ARGSUSED*/
415 xfs_attr_shortform_lookup(xfs_da_args_t *args)
417 xfs_attr_shortform_t *sf;
418 xfs_attr_sf_entry_t *sfe;
419 int i;
420 xfs_ifork_t *ifp;
422 ifp = args->dp->i_afp;
423 ASSERT(ifp->if_flags & XFS_IFINLINE);
424 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
425 sfe = &sf->list[0];
426 for (i = 0; i < sf->hdr.count;
427 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
428 if (sfe->namelen != args->namelen)
429 continue;
430 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
431 continue;
432 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
433 continue;
434 return(XFS_ERROR(EEXIST));
436 return(XFS_ERROR(ENOATTR));
440 * Look up a name in a shortform attribute list structure.
442 /*ARGSUSED*/
444 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
446 xfs_attr_shortform_t *sf;
447 xfs_attr_sf_entry_t *sfe;
448 int i;
450 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
451 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
452 sfe = &sf->list[0];
453 for (i = 0; i < sf->hdr.count;
454 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
455 if (sfe->namelen != args->namelen)
456 continue;
457 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
458 continue;
459 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
460 continue;
461 if (args->flags & ATTR_KERNOVAL) {
462 args->valuelen = sfe->valuelen;
463 return(XFS_ERROR(EEXIST));
465 if (args->valuelen < sfe->valuelen) {
466 args->valuelen = sfe->valuelen;
467 return(XFS_ERROR(ERANGE));
469 args->valuelen = sfe->valuelen;
470 memcpy(args->value, &sfe->nameval[args->namelen],
471 args->valuelen);
472 return(XFS_ERROR(EEXIST));
474 return(XFS_ERROR(ENOATTR));
478 * Convert from using the shortform to the leaf.
481 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
483 xfs_inode_t *dp;
484 xfs_attr_shortform_t *sf;
485 xfs_attr_sf_entry_t *sfe;
486 xfs_da_args_t nargs;
487 char *tmpbuffer;
488 int error, i, size;
489 xfs_dablk_t blkno;
490 xfs_dabuf_t *bp;
491 xfs_ifork_t *ifp;
493 dp = args->dp;
494 ifp = dp->i_afp;
495 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
496 size = be16_to_cpu(sf->hdr.totsize);
497 tmpbuffer = kmem_alloc(size, KM_SLEEP);
498 ASSERT(tmpbuffer != NULL);
499 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
500 sf = (xfs_attr_shortform_t *)tmpbuffer;
502 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
503 bp = NULL;
504 error = xfs_da_grow_inode(args, &blkno);
505 if (error) {
507 * If we hit an IO error middle of the transaction inside
508 * grow_inode(), we may have inconsistent data. Bail out.
510 if (error == EIO)
511 goto out;
512 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
513 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
514 goto out;
517 ASSERT(blkno == 0);
518 error = xfs_attr_leaf_create(args, blkno, &bp);
519 if (error) {
520 error = xfs_da_shrink_inode(args, 0, bp);
521 bp = NULL;
522 if (error)
523 goto out;
524 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
525 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
526 goto out;
529 memset((char *)&nargs, 0, sizeof(nargs));
530 nargs.dp = dp;
531 nargs.firstblock = args->firstblock;
532 nargs.flist = args->flist;
533 nargs.total = args->total;
534 nargs.whichfork = XFS_ATTR_FORK;
535 nargs.trans = args->trans;
536 nargs.oknoent = 1;
538 sfe = &sf->list[0];
539 for (i = 0; i < sf->hdr.count; i++) {
540 nargs.name = (char *)sfe->nameval;
541 nargs.namelen = sfe->namelen;
542 nargs.value = (char *)&sfe->nameval[nargs.namelen];
543 nargs.valuelen = sfe->valuelen;
544 nargs.hashval = xfs_da_hashname((char *)sfe->nameval,
545 sfe->namelen);
546 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
547 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
548 ASSERT(error == ENOATTR);
549 error = xfs_attr_leaf_add(bp, &nargs);
550 ASSERT(error != ENOSPC);
551 if (error)
552 goto out;
553 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
555 error = 0;
557 out:
558 if(bp)
559 xfs_da_buf_done(bp);
560 kmem_free(tmpbuffer, size);
561 return(error);
564 STATIC int
565 xfs_attr_shortform_compare(const void *a, const void *b)
567 xfs_attr_sf_sort_t *sa, *sb;
569 sa = (xfs_attr_sf_sort_t *)a;
570 sb = (xfs_attr_sf_sort_t *)b;
571 if (sa->hash < sb->hash) {
572 return(-1);
573 } else if (sa->hash > sb->hash) {
574 return(1);
575 } else {
576 return(sa->entno - sb->entno);
581 #define XFS_ISRESET_CURSOR(cursor) \
582 (!((cursor)->initted) && !((cursor)->hashval) && \
583 !((cursor)->blkno) && !((cursor)->offset))
585 * Copy out entries of shortform attribute lists for attr_list().
586 * Shortform attribute lists are not stored in hashval sorted order.
587 * If the output buffer is not large enough to hold them all, then we
588 * we have to calculate each entries' hashvalue and sort them before
589 * we can begin returning them to the user.
591 /*ARGSUSED*/
593 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
595 attrlist_cursor_kern_t *cursor;
596 xfs_attr_sf_sort_t *sbuf, *sbp;
597 xfs_attr_shortform_t *sf;
598 xfs_attr_sf_entry_t *sfe;
599 xfs_inode_t *dp;
600 int sbsize, nsbuf, count, i;
601 int error;
603 ASSERT(context != NULL);
604 dp = context->dp;
605 ASSERT(dp != NULL);
606 ASSERT(dp->i_afp != NULL);
607 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
608 ASSERT(sf != NULL);
609 if (!sf->hdr.count)
610 return(0);
611 cursor = context->cursor;
612 ASSERT(cursor != NULL);
614 xfs_attr_trace_l_c("sf start", context);
617 * If the buffer is large enough and the cursor is at the start,
618 * do not bother with sorting since we will return everything in
619 * one buffer and another call using the cursor won't need to be
620 * made.
621 * Note the generous fudge factor of 16 overhead bytes per entry.
622 * If bufsize is zero then put_listent must be a search function
623 * and can just scan through what we have.
625 if (context->bufsize == 0 ||
626 (XFS_ISRESET_CURSOR(cursor) &&
627 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
628 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
629 attrnames_t *namesp;
631 if (!xfs_attr_namesp_match_overrides(context->flags, sfe->flags)) {
632 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
633 continue;
635 namesp = xfs_attr_flags_namesp(sfe->flags);
636 error = context->put_listent(context,
637 namesp,
638 (char *)sfe->nameval,
639 (int)sfe->namelen,
640 (int)sfe->valuelen,
641 (char*)&sfe->nameval[sfe->namelen]);
644 * Either search callback finished early or
645 * didn't fit it all in the buffer after all.
647 if (context->seen_enough)
648 break;
650 if (error)
651 return error;
652 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
654 xfs_attr_trace_l_c("sf big-gulp", context);
655 return(0);
658 /* do no more for a search callback */
659 if (context->bufsize == 0)
660 return 0;
663 * It didn't all fit, so we have to sort everything on hashval.
665 sbsize = sf->hdr.count * sizeof(*sbuf);
666 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
669 * Scan the attribute list for the rest of the entries, storing
670 * the relevant info from only those that match into a buffer.
672 nsbuf = 0;
673 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
674 if (unlikely(
675 ((char *)sfe < (char *)sf) ||
676 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
677 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
678 XFS_ERRLEVEL_LOW,
679 context->dp->i_mount, sfe);
680 xfs_attr_trace_l_c("sf corrupted", context);
681 kmem_free(sbuf, sbsize);
682 return XFS_ERROR(EFSCORRUPTED);
684 if (!xfs_attr_namesp_match_overrides(context->flags, sfe->flags)) {
685 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
686 continue;
688 sbp->entno = i;
689 sbp->hash = xfs_da_hashname((char *)sfe->nameval, sfe->namelen);
690 sbp->name = (char *)sfe->nameval;
691 sbp->namelen = sfe->namelen;
692 /* These are bytes, and both on-disk, don't endian-flip */
693 sbp->valuelen = sfe->valuelen;
694 sbp->flags = sfe->flags;
695 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
696 sbp++;
697 nsbuf++;
701 * Sort the entries on hash then entno.
703 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
706 * Re-find our place IN THE SORTED LIST.
708 count = 0;
709 cursor->initted = 1;
710 cursor->blkno = 0;
711 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
712 if (sbp->hash == cursor->hashval) {
713 if (cursor->offset == count) {
714 break;
716 count++;
717 } else if (sbp->hash > cursor->hashval) {
718 break;
721 if (i == nsbuf) {
722 kmem_free(sbuf, sbsize);
723 xfs_attr_trace_l_c("blk end", context);
724 return(0);
728 * Loop putting entries into the user buffer.
730 for ( ; i < nsbuf; i++, sbp++) {
731 attrnames_t *namesp;
733 namesp = xfs_attr_flags_namesp(sbp->flags);
735 if (cursor->hashval != sbp->hash) {
736 cursor->hashval = sbp->hash;
737 cursor->offset = 0;
739 error = context->put_listent(context,
740 namesp,
741 sbp->name,
742 sbp->namelen,
743 sbp->valuelen,
744 &sbp->name[sbp->namelen]);
745 if (error)
746 return error;
747 if (context->seen_enough)
748 break;
749 cursor->offset++;
752 kmem_free(sbuf, sbsize);
753 xfs_attr_trace_l_c("sf E-O-F", context);
754 return(0);
758 * Check a leaf attribute block to see if all the entries would fit into
759 * a shortform attribute list.
762 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
764 xfs_attr_leafblock_t *leaf;
765 xfs_attr_leaf_entry_t *entry;
766 xfs_attr_leaf_name_local_t *name_loc;
767 int bytes, i;
769 leaf = bp->data;
770 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
772 entry = &leaf->entries[0];
773 bytes = sizeof(struct xfs_attr_sf_hdr);
774 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
775 if (entry->flags & XFS_ATTR_INCOMPLETE)
776 continue; /* don't copy partial entries */
777 if (!(entry->flags & XFS_ATTR_LOCAL))
778 return(0);
779 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
780 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
781 return(0);
782 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
783 return(0);
784 bytes += sizeof(struct xfs_attr_sf_entry)-1
785 + name_loc->namelen
786 + be16_to_cpu(name_loc->valuelen);
788 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
789 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
790 (bytes == sizeof(struct xfs_attr_sf_hdr)))
791 return(-1);
792 return(xfs_attr_shortform_bytesfit(dp, bytes));
796 * Convert a leaf attribute list to shortform attribute list
799 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
801 xfs_attr_leafblock_t *leaf;
802 xfs_attr_leaf_entry_t *entry;
803 xfs_attr_leaf_name_local_t *name_loc;
804 xfs_da_args_t nargs;
805 xfs_inode_t *dp;
806 char *tmpbuffer;
807 int error, i;
809 dp = args->dp;
810 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
811 ASSERT(tmpbuffer != NULL);
813 ASSERT(bp != NULL);
814 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
815 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
816 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
817 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
820 * Clean out the prior contents of the attribute list.
822 error = xfs_da_shrink_inode(args, 0, bp);
823 if (error)
824 goto out;
826 if (forkoff == -1) {
827 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
828 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
831 * Last attribute was removed, revert to original
832 * inode format making all literal area available
833 * to the data fork once more.
835 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
836 dp->i_d.di_forkoff = 0;
837 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
838 ASSERT(dp->i_d.di_anextents == 0);
839 ASSERT(dp->i_afp == NULL);
840 dp->i_df.if_ext_max =
841 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
842 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
843 goto out;
846 xfs_attr_shortform_create(args);
849 * Copy the attributes
851 memset((char *)&nargs, 0, sizeof(nargs));
852 nargs.dp = dp;
853 nargs.firstblock = args->firstblock;
854 nargs.flist = args->flist;
855 nargs.total = args->total;
856 nargs.whichfork = XFS_ATTR_FORK;
857 nargs.trans = args->trans;
858 nargs.oknoent = 1;
859 entry = &leaf->entries[0];
860 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
861 if (entry->flags & XFS_ATTR_INCOMPLETE)
862 continue; /* don't copy partial entries */
863 if (!entry->nameidx)
864 continue;
865 ASSERT(entry->flags & XFS_ATTR_LOCAL);
866 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
867 nargs.name = (char *)name_loc->nameval;
868 nargs.namelen = name_loc->namelen;
869 nargs.value = (char *)&name_loc->nameval[nargs.namelen];
870 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
871 nargs.hashval = be32_to_cpu(entry->hashval);
872 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
873 xfs_attr_shortform_add(&nargs, forkoff);
875 error = 0;
877 out:
878 kmem_free(tmpbuffer, XFS_LBSIZE(dp->i_mount));
879 return(error);
883 * Convert from using a single leaf to a root node and a leaf.
886 xfs_attr_leaf_to_node(xfs_da_args_t *args)
888 xfs_attr_leafblock_t *leaf;
889 xfs_da_intnode_t *node;
890 xfs_inode_t *dp;
891 xfs_dabuf_t *bp1, *bp2;
892 xfs_dablk_t blkno;
893 int error;
895 dp = args->dp;
896 bp1 = bp2 = NULL;
897 error = xfs_da_grow_inode(args, &blkno);
898 if (error)
899 goto out;
900 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
901 XFS_ATTR_FORK);
902 if (error)
903 goto out;
904 ASSERT(bp1 != NULL);
905 bp2 = NULL;
906 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
907 XFS_ATTR_FORK);
908 if (error)
909 goto out;
910 ASSERT(bp2 != NULL);
911 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
912 xfs_da_buf_done(bp1);
913 bp1 = NULL;
914 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
917 * Set up the new root node.
919 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
920 if (error)
921 goto out;
922 node = bp1->data;
923 leaf = bp2->data;
924 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
925 /* both on-disk, don't endian-flip twice */
926 node->btree[0].hashval =
927 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
928 node->btree[0].before = cpu_to_be32(blkno);
929 node->hdr.count = cpu_to_be16(1);
930 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
931 error = 0;
932 out:
933 if (bp1)
934 xfs_da_buf_done(bp1);
935 if (bp2)
936 xfs_da_buf_done(bp2);
937 return(error);
941 /*========================================================================
942 * Routines used for growing the Btree.
943 *========================================================================*/
946 * Create the initial contents of a leaf attribute list
947 * or a leaf in a node attribute list.
949 STATIC int
950 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
952 xfs_attr_leafblock_t *leaf;
953 xfs_attr_leaf_hdr_t *hdr;
954 xfs_inode_t *dp;
955 xfs_dabuf_t *bp;
956 int error;
958 dp = args->dp;
959 ASSERT(dp != NULL);
960 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
961 XFS_ATTR_FORK);
962 if (error)
963 return(error);
964 ASSERT(bp != NULL);
965 leaf = bp->data;
966 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
967 hdr = &leaf->hdr;
968 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
969 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
970 if (!hdr->firstused) {
971 hdr->firstused = cpu_to_be16(
972 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
975 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
976 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
977 sizeof(xfs_attr_leaf_hdr_t));
979 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
981 *bpp = bp;
982 return(0);
986 * Split the leaf node, rebalance, then add the new entry.
989 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
990 xfs_da_state_blk_t *newblk)
992 xfs_dablk_t blkno;
993 int error;
996 * Allocate space for a new leaf node.
998 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
999 error = xfs_da_grow_inode(state->args, &blkno);
1000 if (error)
1001 return(error);
1002 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
1003 if (error)
1004 return(error);
1005 newblk->blkno = blkno;
1006 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1009 * Rebalance the entries across the two leaves.
1010 * NOTE: rebalance() currently depends on the 2nd block being empty.
1012 xfs_attr_leaf_rebalance(state, oldblk, newblk);
1013 error = xfs_da_blk_link(state, oldblk, newblk);
1014 if (error)
1015 return(error);
1018 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1019 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1020 * "new" attrs info. Will need the "old" info to remove it later.
1022 * Insert the "new" entry in the correct block.
1024 if (state->inleaf)
1025 error = xfs_attr_leaf_add(oldblk->bp, state->args);
1026 else
1027 error = xfs_attr_leaf_add(newblk->bp, state->args);
1030 * Update last hashval in each block since we added the name.
1032 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1033 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1034 return(error);
1038 * Add a name to the leaf attribute list structure.
1041 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
1043 xfs_attr_leafblock_t *leaf;
1044 xfs_attr_leaf_hdr_t *hdr;
1045 xfs_attr_leaf_map_t *map;
1046 int tablesize, entsize, sum, tmp, i;
1048 leaf = bp->data;
1049 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1050 ASSERT((args->index >= 0)
1051 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1052 hdr = &leaf->hdr;
1053 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1054 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1057 * Search through freemap for first-fit on new name length.
1058 * (may need to figure in size of entry struct too)
1060 tablesize = (be16_to_cpu(hdr->count) + 1)
1061 * sizeof(xfs_attr_leaf_entry_t)
1062 + sizeof(xfs_attr_leaf_hdr_t);
1063 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1064 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1065 if (tablesize > be16_to_cpu(hdr->firstused)) {
1066 sum += be16_to_cpu(map->size);
1067 continue;
1069 if (!map->size)
1070 continue; /* no space in this map */
1071 tmp = entsize;
1072 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1073 tmp += sizeof(xfs_attr_leaf_entry_t);
1074 if (be16_to_cpu(map->size) >= tmp) {
1075 tmp = xfs_attr_leaf_add_work(bp, args, i);
1076 return(tmp);
1078 sum += be16_to_cpu(map->size);
1082 * If there are no holes in the address space of the block,
1083 * and we don't have enough freespace, then compaction will do us
1084 * no good and we should just give up.
1086 if (!hdr->holes && (sum < entsize))
1087 return(XFS_ERROR(ENOSPC));
1090 * Compact the entries to coalesce free space.
1091 * This may change the hdr->count via dropping INCOMPLETE entries.
1093 xfs_attr_leaf_compact(args->trans, bp);
1096 * After compaction, the block is guaranteed to have only one
1097 * free region, in freemap[0]. If it is not big enough, give up.
1099 if (be16_to_cpu(hdr->freemap[0].size)
1100 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1101 return(XFS_ERROR(ENOSPC));
1103 return(xfs_attr_leaf_add_work(bp, args, 0));
1107 * Add a name to a leaf attribute list structure.
1109 STATIC int
1110 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1112 xfs_attr_leafblock_t *leaf;
1113 xfs_attr_leaf_hdr_t *hdr;
1114 xfs_attr_leaf_entry_t *entry;
1115 xfs_attr_leaf_name_local_t *name_loc;
1116 xfs_attr_leaf_name_remote_t *name_rmt;
1117 xfs_attr_leaf_map_t *map;
1118 xfs_mount_t *mp;
1119 int tmp, i;
1121 leaf = bp->data;
1122 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1123 hdr = &leaf->hdr;
1124 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1125 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1128 * Force open some space in the entry array and fill it in.
1130 entry = &leaf->entries[args->index];
1131 if (args->index < be16_to_cpu(hdr->count)) {
1132 tmp = be16_to_cpu(hdr->count) - args->index;
1133 tmp *= sizeof(xfs_attr_leaf_entry_t);
1134 memmove((char *)(entry+1), (char *)entry, tmp);
1135 xfs_da_log_buf(args->trans, bp,
1136 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1138 be16_add(&hdr->count, 1);
1141 * Allocate space for the new string (at the end of the run).
1143 map = &hdr->freemap[mapindex];
1144 mp = args->trans->t_mountp;
1145 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1146 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1147 ASSERT(be16_to_cpu(map->size) >=
1148 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1149 mp->m_sb.sb_blocksize, NULL));
1150 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1151 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1152 be16_add(&map->size,
1153 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1154 mp->m_sb.sb_blocksize, &tmp));
1155 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1156 be16_to_cpu(map->size));
1157 entry->hashval = cpu_to_be32(args->hashval);
1158 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1159 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1160 if (args->rename) {
1161 entry->flags |= XFS_ATTR_INCOMPLETE;
1162 if ((args->blkno2 == args->blkno) &&
1163 (args->index2 <= args->index)) {
1164 args->index2++;
1167 xfs_da_log_buf(args->trans, bp,
1168 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1169 ASSERT((args->index == 0) ||
1170 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1171 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1172 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1175 * Copy the attribute name and value into the new space.
1177 * For "remote" attribute values, simply note that we need to
1178 * allocate space for the "remote" value. We can't actually
1179 * allocate the extents in this transaction, and we can't decide
1180 * which blocks they should be as we might allocate more blocks
1181 * as part of this transaction (a split operation for example).
1183 if (entry->flags & XFS_ATTR_LOCAL) {
1184 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1185 name_loc->namelen = args->namelen;
1186 name_loc->valuelen = cpu_to_be16(args->valuelen);
1187 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1188 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1189 be16_to_cpu(name_loc->valuelen));
1190 } else {
1191 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1192 name_rmt->namelen = args->namelen;
1193 memcpy((char *)name_rmt->name, args->name, args->namelen);
1194 entry->flags |= XFS_ATTR_INCOMPLETE;
1195 /* just in case */
1196 name_rmt->valuelen = 0;
1197 name_rmt->valueblk = 0;
1198 args->rmtblkno = 1;
1199 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1201 xfs_da_log_buf(args->trans, bp,
1202 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1203 xfs_attr_leaf_entsize(leaf, args->index)));
1206 * Update the control info for this leaf node
1208 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1209 /* both on-disk, don't endian-flip twice */
1210 hdr->firstused = entry->nameidx;
1212 ASSERT(be16_to_cpu(hdr->firstused) >=
1213 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1214 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1215 + sizeof(xfs_attr_leaf_hdr_t);
1216 map = &hdr->freemap[0];
1217 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1218 if (be16_to_cpu(map->base) == tmp) {
1219 be16_add(&map->base, sizeof(xfs_attr_leaf_entry_t));
1220 be16_add(&map->size,
1221 -((int)sizeof(xfs_attr_leaf_entry_t)));
1224 be16_add(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1225 xfs_da_log_buf(args->trans, bp,
1226 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1227 return(0);
1231 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1233 STATIC void
1234 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1236 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1237 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1238 xfs_mount_t *mp;
1239 char *tmpbuffer;
1241 mp = trans->t_mountp;
1242 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1243 ASSERT(tmpbuffer != NULL);
1244 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1245 memset(bp->data, 0, XFS_LBSIZE(mp));
1248 * Copy basic information
1250 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1251 leaf_d = bp->data;
1252 hdr_s = &leaf_s->hdr;
1253 hdr_d = &leaf_d->hdr;
1254 hdr_d->info = hdr_s->info; /* struct copy */
1255 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1256 /* handle truncation gracefully */
1257 if (!hdr_d->firstused) {
1258 hdr_d->firstused = cpu_to_be16(
1259 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1261 hdr_d->usedbytes = 0;
1262 hdr_d->count = 0;
1263 hdr_d->holes = 0;
1264 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1265 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1266 sizeof(xfs_attr_leaf_hdr_t));
1269 * Copy all entry's in the same (sorted) order,
1270 * but allocate name/value pairs packed and in sequence.
1272 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1273 be16_to_cpu(hdr_s->count), mp);
1274 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1276 kmem_free(tmpbuffer, XFS_LBSIZE(mp));
1280 * Redistribute the attribute list entries between two leaf nodes,
1281 * taking into account the size of the new entry.
1283 * NOTE: if new block is empty, then it will get the upper half of the
1284 * old block. At present, all (one) callers pass in an empty second block.
1286 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1287 * to match what it is doing in splitting the attribute leaf block. Those
1288 * values are used in "atomic rename" operations on attributes. Note that
1289 * the "new" and "old" values can end up in different blocks.
1291 STATIC void
1292 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1293 xfs_da_state_blk_t *blk2)
1295 xfs_da_args_t *args;
1296 xfs_da_state_blk_t *tmp_blk;
1297 xfs_attr_leafblock_t *leaf1, *leaf2;
1298 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1299 int count, totallen, max, space, swap;
1302 * Set up environment.
1304 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1305 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1306 leaf1 = blk1->bp->data;
1307 leaf2 = blk2->bp->data;
1308 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1309 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1310 args = state->args;
1313 * Check ordering of blocks, reverse if it makes things simpler.
1315 * NOTE: Given that all (current) callers pass in an empty
1316 * second block, this code should never set "swap".
1318 swap = 0;
1319 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1320 tmp_blk = blk1;
1321 blk1 = blk2;
1322 blk2 = tmp_blk;
1323 leaf1 = blk1->bp->data;
1324 leaf2 = blk2->bp->data;
1325 swap = 1;
1327 hdr1 = &leaf1->hdr;
1328 hdr2 = &leaf2->hdr;
1331 * Examine entries until we reduce the absolute difference in
1332 * byte usage between the two blocks to a minimum. Then get
1333 * the direction to copy and the number of elements to move.
1335 * "inleaf" is true if the new entry should be inserted into blk1.
1336 * If "swap" is also true, then reverse the sense of "inleaf".
1338 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1339 &count, &totallen);
1340 if (swap)
1341 state->inleaf = !state->inleaf;
1344 * Move any entries required from leaf to leaf:
1346 if (count < be16_to_cpu(hdr1->count)) {
1348 * Figure the total bytes to be added to the destination leaf.
1350 /* number entries being moved */
1351 count = be16_to_cpu(hdr1->count) - count;
1352 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1353 space += count * sizeof(xfs_attr_leaf_entry_t);
1356 * leaf2 is the destination, compact it if it looks tight.
1358 max = be16_to_cpu(hdr2->firstused)
1359 - sizeof(xfs_attr_leaf_hdr_t);
1360 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1361 if (space > max) {
1362 xfs_attr_leaf_compact(args->trans, blk2->bp);
1366 * Move high entries from leaf1 to low end of leaf2.
1368 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1369 leaf2, 0, count, state->mp);
1371 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1372 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1373 } else if (count > be16_to_cpu(hdr1->count)) {
1375 * I assert that since all callers pass in an empty
1376 * second buffer, this code should never execute.
1380 * Figure the total bytes to be added to the destination leaf.
1382 /* number entries being moved */
1383 count -= be16_to_cpu(hdr1->count);
1384 space = totallen - be16_to_cpu(hdr1->usedbytes);
1385 space += count * sizeof(xfs_attr_leaf_entry_t);
1388 * leaf1 is the destination, compact it if it looks tight.
1390 max = be16_to_cpu(hdr1->firstused)
1391 - sizeof(xfs_attr_leaf_hdr_t);
1392 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1393 if (space > max) {
1394 xfs_attr_leaf_compact(args->trans, blk1->bp);
1398 * Move low entries from leaf2 to high end of leaf1.
1400 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1401 be16_to_cpu(hdr1->count), count, state->mp);
1403 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1404 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1408 * Copy out last hashval in each block for B-tree code.
1410 blk1->hashval = be32_to_cpu(
1411 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1412 blk2->hashval = be32_to_cpu(
1413 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1416 * Adjust the expected index for insertion.
1417 * NOTE: this code depends on the (current) situation that the
1418 * second block was originally empty.
1420 * If the insertion point moved to the 2nd block, we must adjust
1421 * the index. We must also track the entry just following the
1422 * new entry for use in an "atomic rename" operation, that entry
1423 * is always the "old" entry and the "new" entry is what we are
1424 * inserting. The index/blkno fields refer to the "old" entry,
1425 * while the index2/blkno2 fields refer to the "new" entry.
1427 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1428 ASSERT(state->inleaf == 0);
1429 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1430 args->index = args->index2 = blk2->index;
1431 args->blkno = args->blkno2 = blk2->blkno;
1432 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1433 if (state->inleaf) {
1434 args->index = blk1->index;
1435 args->blkno = blk1->blkno;
1436 args->index2 = 0;
1437 args->blkno2 = blk2->blkno;
1438 } else {
1439 blk2->index = blk1->index
1440 - be16_to_cpu(leaf1->hdr.count);
1441 args->index = args->index2 = blk2->index;
1442 args->blkno = args->blkno2 = blk2->blkno;
1444 } else {
1445 ASSERT(state->inleaf == 1);
1446 args->index = args->index2 = blk1->index;
1447 args->blkno = args->blkno2 = blk1->blkno;
1452 * Examine entries until we reduce the absolute difference in
1453 * byte usage between the two blocks to a minimum.
1454 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1455 * GROT: there will always be enough room in either block for a new entry.
1456 * GROT: Do a double-split for this case?
1458 STATIC int
1459 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1460 xfs_da_state_blk_t *blk1,
1461 xfs_da_state_blk_t *blk2,
1462 int *countarg, int *usedbytesarg)
1464 xfs_attr_leafblock_t *leaf1, *leaf2;
1465 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1466 xfs_attr_leaf_entry_t *entry;
1467 int count, max, index, totallen, half;
1468 int lastdelta, foundit, tmp;
1471 * Set up environment.
1473 leaf1 = blk1->bp->data;
1474 leaf2 = blk2->bp->data;
1475 hdr1 = &leaf1->hdr;
1476 hdr2 = &leaf2->hdr;
1477 foundit = 0;
1478 totallen = 0;
1481 * Examine entries until we reduce the absolute difference in
1482 * byte usage between the two blocks to a minimum.
1484 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1485 half = (max+1) * sizeof(*entry);
1486 half += be16_to_cpu(hdr1->usedbytes) +
1487 be16_to_cpu(hdr2->usedbytes) +
1488 xfs_attr_leaf_newentsize(
1489 state->args->namelen,
1490 state->args->valuelen,
1491 state->blocksize, NULL);
1492 half /= 2;
1493 lastdelta = state->blocksize;
1494 entry = &leaf1->entries[0];
1495 for (count = index = 0; count < max; entry++, index++, count++) {
1497 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1499 * The new entry is in the first block, account for it.
1501 if (count == blk1->index) {
1502 tmp = totallen + sizeof(*entry) +
1503 xfs_attr_leaf_newentsize(
1504 state->args->namelen,
1505 state->args->valuelen,
1506 state->blocksize, NULL);
1507 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1508 break;
1509 lastdelta = XFS_ATTR_ABS(half - tmp);
1510 totallen = tmp;
1511 foundit = 1;
1515 * Wrap around into the second block if necessary.
1517 if (count == be16_to_cpu(hdr1->count)) {
1518 leaf1 = leaf2;
1519 entry = &leaf1->entries[0];
1520 index = 0;
1524 * Figure out if next leaf entry would be too much.
1526 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1527 index);
1528 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1529 break;
1530 lastdelta = XFS_ATTR_ABS(half - tmp);
1531 totallen = tmp;
1532 #undef XFS_ATTR_ABS
1536 * Calculate the number of usedbytes that will end up in lower block.
1537 * If new entry not in lower block, fix up the count.
1539 totallen -= count * sizeof(*entry);
1540 if (foundit) {
1541 totallen -= sizeof(*entry) +
1542 xfs_attr_leaf_newentsize(
1543 state->args->namelen,
1544 state->args->valuelen,
1545 state->blocksize, NULL);
1548 *countarg = count;
1549 *usedbytesarg = totallen;
1550 return(foundit);
1553 /*========================================================================
1554 * Routines used for shrinking the Btree.
1555 *========================================================================*/
1558 * Check a leaf block and its neighbors to see if the block should be
1559 * collapsed into one or the other neighbor. Always keep the block
1560 * with the smaller block number.
1561 * If the current block is over 50% full, don't try to join it, return 0.
1562 * If the block is empty, fill in the state structure and return 2.
1563 * If it can be collapsed, fill in the state structure and return 1.
1564 * If nothing can be done, return 0.
1566 * GROT: allow for INCOMPLETE entries in calculation.
1569 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1571 xfs_attr_leafblock_t *leaf;
1572 xfs_da_state_blk_t *blk;
1573 xfs_da_blkinfo_t *info;
1574 int count, bytes, forward, error, retval, i;
1575 xfs_dablk_t blkno;
1576 xfs_dabuf_t *bp;
1579 * Check for the degenerate case of the block being over 50% full.
1580 * If so, it's not worth even looking to see if we might be able
1581 * to coalesce with a sibling.
1583 blk = &state->path.blk[ state->path.active-1 ];
1584 info = blk->bp->data;
1585 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1586 leaf = (xfs_attr_leafblock_t *)info;
1587 count = be16_to_cpu(leaf->hdr.count);
1588 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1589 count * sizeof(xfs_attr_leaf_entry_t) +
1590 be16_to_cpu(leaf->hdr.usedbytes);
1591 if (bytes > (state->blocksize >> 1)) {
1592 *action = 0; /* blk over 50%, don't try to join */
1593 return(0);
1597 * Check for the degenerate case of the block being empty.
1598 * If the block is empty, we'll simply delete it, no need to
1599 * coalesce it with a sibling block. We choose (arbitrarily)
1600 * to merge with the forward block unless it is NULL.
1602 if (count == 0) {
1604 * Make altpath point to the block we want to keep and
1605 * path point to the block we want to drop (this one).
1607 forward = (info->forw != 0);
1608 memcpy(&state->altpath, &state->path, sizeof(state->path));
1609 error = xfs_da_path_shift(state, &state->altpath, forward,
1610 0, &retval);
1611 if (error)
1612 return(error);
1613 if (retval) {
1614 *action = 0;
1615 } else {
1616 *action = 2;
1618 return(0);
1622 * Examine each sibling block to see if we can coalesce with
1623 * at least 25% free space to spare. We need to figure out
1624 * whether to merge with the forward or the backward block.
1625 * We prefer coalescing with the lower numbered sibling so as
1626 * to shrink an attribute list over time.
1628 /* start with smaller blk num */
1629 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1630 for (i = 0; i < 2; forward = !forward, i++) {
1631 if (forward)
1632 blkno = be32_to_cpu(info->forw);
1633 else
1634 blkno = be32_to_cpu(info->back);
1635 if (blkno == 0)
1636 continue;
1637 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1638 blkno, -1, &bp, XFS_ATTR_FORK);
1639 if (error)
1640 return(error);
1641 ASSERT(bp != NULL);
1643 leaf = (xfs_attr_leafblock_t *)info;
1644 count = be16_to_cpu(leaf->hdr.count);
1645 bytes = state->blocksize - (state->blocksize>>2);
1646 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1647 leaf = bp->data;
1648 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1649 count += be16_to_cpu(leaf->hdr.count);
1650 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1651 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1652 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1653 xfs_da_brelse(state->args->trans, bp);
1654 if (bytes >= 0)
1655 break; /* fits with at least 25% to spare */
1657 if (i >= 2) {
1658 *action = 0;
1659 return(0);
1663 * Make altpath point to the block we want to keep (the lower
1664 * numbered block) and path point to the block we want to drop.
1666 memcpy(&state->altpath, &state->path, sizeof(state->path));
1667 if (blkno < blk->blkno) {
1668 error = xfs_da_path_shift(state, &state->altpath, forward,
1669 0, &retval);
1670 } else {
1671 error = xfs_da_path_shift(state, &state->path, forward,
1672 0, &retval);
1674 if (error)
1675 return(error);
1676 if (retval) {
1677 *action = 0;
1678 } else {
1679 *action = 1;
1681 return(0);
1685 * Remove a name from the leaf attribute list structure.
1687 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1688 * If two leaves are 37% full, when combined they will leave 25% free.
1691 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1693 xfs_attr_leafblock_t *leaf;
1694 xfs_attr_leaf_hdr_t *hdr;
1695 xfs_attr_leaf_map_t *map;
1696 xfs_attr_leaf_entry_t *entry;
1697 int before, after, smallest, entsize;
1698 int tablesize, tmp, i;
1699 xfs_mount_t *mp;
1701 leaf = bp->data;
1702 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1703 hdr = &leaf->hdr;
1704 mp = args->trans->t_mountp;
1705 ASSERT((be16_to_cpu(hdr->count) > 0)
1706 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1707 ASSERT((args->index >= 0)
1708 && (args->index < be16_to_cpu(hdr->count)));
1709 ASSERT(be16_to_cpu(hdr->firstused) >=
1710 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1711 entry = &leaf->entries[args->index];
1712 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1713 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1716 * Scan through free region table:
1717 * check for adjacency of free'd entry with an existing one,
1718 * find smallest free region in case we need to replace it,
1719 * adjust any map that borders the entry table,
1721 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1722 + sizeof(xfs_attr_leaf_hdr_t);
1723 map = &hdr->freemap[0];
1724 tmp = be16_to_cpu(map->size);
1725 before = after = -1;
1726 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1727 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1728 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1729 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1730 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1731 if (be16_to_cpu(map->base) == tablesize) {
1732 be16_add(&map->base,
1733 -((int)sizeof(xfs_attr_leaf_entry_t)));
1734 be16_add(&map->size, sizeof(xfs_attr_leaf_entry_t));
1737 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1738 == be16_to_cpu(entry->nameidx)) {
1739 before = i;
1740 } else if (be16_to_cpu(map->base)
1741 == (be16_to_cpu(entry->nameidx) + entsize)) {
1742 after = i;
1743 } else if (be16_to_cpu(map->size) < tmp) {
1744 tmp = be16_to_cpu(map->size);
1745 smallest = i;
1750 * Coalesce adjacent freemap regions,
1751 * or replace the smallest region.
1753 if ((before >= 0) || (after >= 0)) {
1754 if ((before >= 0) && (after >= 0)) {
1755 map = &hdr->freemap[before];
1756 be16_add(&map->size, entsize);
1757 be16_add(&map->size,
1758 be16_to_cpu(hdr->freemap[after].size));
1759 hdr->freemap[after].base = 0;
1760 hdr->freemap[after].size = 0;
1761 } else if (before >= 0) {
1762 map = &hdr->freemap[before];
1763 be16_add(&map->size, entsize);
1764 } else {
1765 map = &hdr->freemap[after];
1766 /* both on-disk, don't endian flip twice */
1767 map->base = entry->nameidx;
1768 be16_add(&map->size, entsize);
1770 } else {
1772 * Replace smallest region (if it is smaller than free'd entry)
1774 map = &hdr->freemap[smallest];
1775 if (be16_to_cpu(map->size) < entsize) {
1776 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1777 map->size = cpu_to_be16(entsize);
1782 * Did we remove the first entry?
1784 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1785 smallest = 1;
1786 else
1787 smallest = 0;
1790 * Compress the remaining entries and zero out the removed stuff.
1792 memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1793 be16_add(&hdr->usedbytes, -entsize);
1794 xfs_da_log_buf(args->trans, bp,
1795 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1796 entsize));
1798 tmp = (be16_to_cpu(hdr->count) - args->index)
1799 * sizeof(xfs_attr_leaf_entry_t);
1800 memmove((char *)entry, (char *)(entry+1), tmp);
1801 be16_add(&hdr->count, -1);
1802 xfs_da_log_buf(args->trans, bp,
1803 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1804 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1805 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1808 * If we removed the first entry, re-find the first used byte
1809 * in the name area. Note that if the entry was the "firstused",
1810 * then we don't have a "hole" in our block resulting from
1811 * removing the name.
1813 if (smallest) {
1814 tmp = XFS_LBSIZE(mp);
1815 entry = &leaf->entries[0];
1816 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1817 ASSERT(be16_to_cpu(entry->nameidx) >=
1818 be16_to_cpu(hdr->firstused));
1819 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1821 if (be16_to_cpu(entry->nameidx) < tmp)
1822 tmp = be16_to_cpu(entry->nameidx);
1824 hdr->firstused = cpu_to_be16(tmp);
1825 if (!hdr->firstused) {
1826 hdr->firstused = cpu_to_be16(
1827 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1829 } else {
1830 hdr->holes = 1; /* mark as needing compaction */
1832 xfs_da_log_buf(args->trans, bp,
1833 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1836 * Check if leaf is less than 50% full, caller may want to
1837 * "join" the leaf with a sibling if so.
1839 tmp = sizeof(xfs_attr_leaf_hdr_t);
1840 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1841 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1842 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1846 * Move all the attribute list entries from drop_leaf into save_leaf.
1848 void
1849 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1850 xfs_da_state_blk_t *save_blk)
1852 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1853 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1854 xfs_mount_t *mp;
1855 char *tmpbuffer;
1858 * Set up environment.
1860 mp = state->mp;
1861 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1862 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1863 drop_leaf = drop_blk->bp->data;
1864 save_leaf = save_blk->bp->data;
1865 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1866 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1867 drop_hdr = &drop_leaf->hdr;
1868 save_hdr = &save_leaf->hdr;
1871 * Save last hashval from dying block for later Btree fixup.
1873 drop_blk->hashval = be32_to_cpu(
1874 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1877 * Check if we need a temp buffer, or can we do it in place.
1878 * Note that we don't check "leaf" for holes because we will
1879 * always be dropping it, toosmall() decided that for us already.
1881 if (save_hdr->holes == 0) {
1883 * dest leaf has no holes, so we add there. May need
1884 * to make some room in the entry array.
1886 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1887 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1888 be16_to_cpu(drop_hdr->count), mp);
1889 } else {
1890 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1891 be16_to_cpu(save_hdr->count),
1892 be16_to_cpu(drop_hdr->count), mp);
1894 } else {
1896 * Destination has holes, so we make a temporary copy
1897 * of the leaf and add them both to that.
1899 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1900 ASSERT(tmpbuffer != NULL);
1901 memset(tmpbuffer, 0, state->blocksize);
1902 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1903 tmp_hdr = &tmp_leaf->hdr;
1904 tmp_hdr->info = save_hdr->info; /* struct copy */
1905 tmp_hdr->count = 0;
1906 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1907 if (!tmp_hdr->firstused) {
1908 tmp_hdr->firstused = cpu_to_be16(
1909 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1911 tmp_hdr->usedbytes = 0;
1912 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1913 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1914 be16_to_cpu(drop_hdr->count), mp);
1915 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1916 be16_to_cpu(tmp_leaf->hdr.count),
1917 be16_to_cpu(save_hdr->count), mp);
1918 } else {
1919 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1920 be16_to_cpu(save_hdr->count), mp);
1921 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1922 be16_to_cpu(tmp_leaf->hdr.count),
1923 be16_to_cpu(drop_hdr->count), mp);
1925 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1926 kmem_free(tmpbuffer, state->blocksize);
1929 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1930 state->blocksize - 1);
1933 * Copy out last hashval in each block for B-tree code.
1935 save_blk->hashval = be32_to_cpu(
1936 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1939 /*========================================================================
1940 * Routines used for finding things in the Btree.
1941 *========================================================================*/
1944 * Look up a name in a leaf attribute list structure.
1945 * This is the internal routine, it uses the caller's buffer.
1947 * Note that duplicate keys are allowed, but only check within the
1948 * current leaf node. The Btree code must check in adjacent leaf nodes.
1950 * Return in args->index the index into the entry[] array of either
1951 * the found entry, or where the entry should have been (insert before
1952 * that entry).
1954 * Don't change the args->value unless we find the attribute.
1957 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1959 xfs_attr_leafblock_t *leaf;
1960 xfs_attr_leaf_entry_t *entry;
1961 xfs_attr_leaf_name_local_t *name_loc;
1962 xfs_attr_leaf_name_remote_t *name_rmt;
1963 int probe, span;
1964 xfs_dahash_t hashval;
1966 leaf = bp->data;
1967 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1968 ASSERT(be16_to_cpu(leaf->hdr.count)
1969 < (XFS_LBSIZE(args->dp->i_mount)/8));
1972 * Binary search. (note: small blocks will skip this loop)
1974 hashval = args->hashval;
1975 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1976 for (entry = &leaf->entries[probe]; span > 4;
1977 entry = &leaf->entries[probe]) {
1978 span /= 2;
1979 if (be32_to_cpu(entry->hashval) < hashval)
1980 probe += span;
1981 else if (be32_to_cpu(entry->hashval) > hashval)
1982 probe -= span;
1983 else
1984 break;
1986 ASSERT((probe >= 0) &&
1987 (!leaf->hdr.count
1988 || (probe < be16_to_cpu(leaf->hdr.count))));
1989 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1992 * Since we may have duplicate hashval's, find the first matching
1993 * hashval in the leaf.
1995 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1996 entry--;
1997 probe--;
1999 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
2000 (be32_to_cpu(entry->hashval) < hashval)) {
2001 entry++;
2002 probe++;
2004 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
2005 (be32_to_cpu(entry->hashval) != hashval)) {
2006 args->index = probe;
2007 return(XFS_ERROR(ENOATTR));
2011 * Duplicate keys may be present, so search all of them for a match.
2013 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
2014 (be32_to_cpu(entry->hashval) == hashval);
2015 entry++, probe++) {
2017 * GROT: Add code to remove incomplete entries.
2020 * If we are looking for INCOMPLETE entries, show only those.
2021 * If we are looking for complete entries, show only those.
2023 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2024 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2025 continue;
2027 if (entry->flags & XFS_ATTR_LOCAL) {
2028 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
2029 if (name_loc->namelen != args->namelen)
2030 continue;
2031 if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
2032 continue;
2033 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2034 continue;
2035 args->index = probe;
2036 return(XFS_ERROR(EEXIST));
2037 } else {
2038 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
2039 if (name_rmt->namelen != args->namelen)
2040 continue;
2041 if (memcmp(args->name, (char *)name_rmt->name,
2042 args->namelen) != 0)
2043 continue;
2044 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2045 continue;
2046 args->index = probe;
2047 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2048 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2049 be32_to_cpu(name_rmt->valuelen));
2050 return(XFS_ERROR(EEXIST));
2053 args->index = probe;
2054 return(XFS_ERROR(ENOATTR));
2058 * Get the value associated with an attribute name from a leaf attribute
2059 * list structure.
2062 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2064 int valuelen;
2065 xfs_attr_leafblock_t *leaf;
2066 xfs_attr_leaf_entry_t *entry;
2067 xfs_attr_leaf_name_local_t *name_loc;
2068 xfs_attr_leaf_name_remote_t *name_rmt;
2070 leaf = bp->data;
2071 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2072 ASSERT(be16_to_cpu(leaf->hdr.count)
2073 < (XFS_LBSIZE(args->dp->i_mount)/8));
2074 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2076 entry = &leaf->entries[args->index];
2077 if (entry->flags & XFS_ATTR_LOCAL) {
2078 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2079 ASSERT(name_loc->namelen == args->namelen);
2080 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2081 valuelen = be16_to_cpu(name_loc->valuelen);
2082 if (args->flags & ATTR_KERNOVAL) {
2083 args->valuelen = valuelen;
2084 return(0);
2086 if (args->valuelen < valuelen) {
2087 args->valuelen = valuelen;
2088 return(XFS_ERROR(ERANGE));
2090 args->valuelen = valuelen;
2091 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2092 } else {
2093 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2094 ASSERT(name_rmt->namelen == args->namelen);
2095 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2096 valuelen = be32_to_cpu(name_rmt->valuelen);
2097 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2098 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2099 if (args->flags & ATTR_KERNOVAL) {
2100 args->valuelen = valuelen;
2101 return(0);
2103 if (args->valuelen < valuelen) {
2104 args->valuelen = valuelen;
2105 return(XFS_ERROR(ERANGE));
2107 args->valuelen = valuelen;
2109 return(0);
2112 /*========================================================================
2113 * Utility routines.
2114 *========================================================================*/
2117 * Move the indicated entries from one leaf to another.
2118 * NOTE: this routine modifies both source and destination leaves.
2120 /*ARGSUSED*/
2121 STATIC void
2122 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2123 xfs_attr_leafblock_t *leaf_d, int start_d,
2124 int count, xfs_mount_t *mp)
2126 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2127 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2128 int desti, tmp, i;
2131 * Check for nothing to do.
2133 if (count == 0)
2134 return;
2137 * Set up environment.
2139 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2140 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2141 hdr_s = &leaf_s->hdr;
2142 hdr_d = &leaf_d->hdr;
2143 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2144 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2145 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2146 ((be16_to_cpu(hdr_s->count)
2147 * sizeof(*entry_s))+sizeof(*hdr_s)));
2148 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2149 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2150 ((be16_to_cpu(hdr_d->count)
2151 * sizeof(*entry_d))+sizeof(*hdr_d)));
2153 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2154 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2155 ASSERT(count <= be16_to_cpu(hdr_s->count));
2158 * Move the entries in the destination leaf up to make a hole?
2160 if (start_d < be16_to_cpu(hdr_d->count)) {
2161 tmp = be16_to_cpu(hdr_d->count) - start_d;
2162 tmp *= sizeof(xfs_attr_leaf_entry_t);
2163 entry_s = &leaf_d->entries[start_d];
2164 entry_d = &leaf_d->entries[start_d + count];
2165 memmove((char *)entry_d, (char *)entry_s, tmp);
2169 * Copy all entry's in the same (sorted) order,
2170 * but allocate attribute info packed and in sequence.
2172 entry_s = &leaf_s->entries[start_s];
2173 entry_d = &leaf_d->entries[start_d];
2174 desti = start_d;
2175 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2176 ASSERT(be16_to_cpu(entry_s->nameidx)
2177 >= be16_to_cpu(hdr_s->firstused));
2178 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2179 #ifdef GROT
2181 * Code to drop INCOMPLETE entries. Difficult to use as we
2182 * may also need to change the insertion index. Code turned
2183 * off for 6.2, should be revisited later.
2185 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2186 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2187 be16_add(&hdr_s->usedbytes, -tmp);
2188 be16_add(&hdr_s->count, -1);
2189 entry_d--; /* to compensate for ++ in loop hdr */
2190 desti--;
2191 if ((start_s + i) < offset)
2192 result++; /* insertion index adjustment */
2193 } else {
2194 #endif /* GROT */
2195 be16_add(&hdr_d->firstused, -tmp);
2196 /* both on-disk, don't endian flip twice */
2197 entry_d->hashval = entry_s->hashval;
2198 /* both on-disk, don't endian flip twice */
2199 entry_d->nameidx = hdr_d->firstused;
2200 entry_d->flags = entry_s->flags;
2201 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2202 <= XFS_LBSIZE(mp));
2203 memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2204 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2205 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2206 <= XFS_LBSIZE(mp));
2207 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2208 be16_add(&hdr_s->usedbytes, -tmp);
2209 be16_add(&hdr_d->usedbytes, tmp);
2210 be16_add(&hdr_s->count, -1);
2211 be16_add(&hdr_d->count, 1);
2212 tmp = be16_to_cpu(hdr_d->count)
2213 * sizeof(xfs_attr_leaf_entry_t)
2214 + sizeof(xfs_attr_leaf_hdr_t);
2215 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2216 #ifdef GROT
2218 #endif /* GROT */
2222 * Zero out the entries we just copied.
2224 if (start_s == be16_to_cpu(hdr_s->count)) {
2225 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2226 entry_s = &leaf_s->entries[start_s];
2227 ASSERT(((char *)entry_s + tmp) <=
2228 ((char *)leaf_s + XFS_LBSIZE(mp)));
2229 memset((char *)entry_s, 0, tmp);
2230 } else {
2232 * Move the remaining entries down to fill the hole,
2233 * then zero the entries at the top.
2235 tmp = be16_to_cpu(hdr_s->count) - count;
2236 tmp *= sizeof(xfs_attr_leaf_entry_t);
2237 entry_s = &leaf_s->entries[start_s + count];
2238 entry_d = &leaf_s->entries[start_s];
2239 memmove((char *)entry_d, (char *)entry_s, tmp);
2241 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2242 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2243 ASSERT(((char *)entry_s + tmp) <=
2244 ((char *)leaf_s + XFS_LBSIZE(mp)));
2245 memset((char *)entry_s, 0, tmp);
2249 * Fill in the freemap information
2251 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2252 be16_add(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2253 sizeof(xfs_attr_leaf_entry_t));
2254 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2255 - be16_to_cpu(hdr_d->freemap[0].base));
2256 hdr_d->freemap[1].base = 0;
2257 hdr_d->freemap[2].base = 0;
2258 hdr_d->freemap[1].size = 0;
2259 hdr_d->freemap[2].size = 0;
2260 hdr_s->holes = 1; /* leaf may not be compact */
2264 * Compare two leaf blocks "order".
2265 * Return 0 unless leaf2 should go before leaf1.
2268 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2270 xfs_attr_leafblock_t *leaf1, *leaf2;
2272 leaf1 = leaf1_bp->data;
2273 leaf2 = leaf2_bp->data;
2274 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2275 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2276 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2277 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2278 ((be32_to_cpu(leaf2->entries[0].hashval) <
2279 be32_to_cpu(leaf1->entries[0].hashval)) ||
2280 (be32_to_cpu(leaf2->entries[
2281 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2282 be32_to_cpu(leaf1->entries[
2283 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2284 return(1);
2286 return(0);
2290 * Pick up the last hashvalue from a leaf block.
2292 xfs_dahash_t
2293 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2295 xfs_attr_leafblock_t *leaf;
2297 leaf = bp->data;
2298 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2299 if (count)
2300 *count = be16_to_cpu(leaf->hdr.count);
2301 if (!leaf->hdr.count)
2302 return(0);
2303 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2307 * Calculate the number of bytes used to store the indicated attribute
2308 * (whether local or remote only calculate bytes in this block).
2310 STATIC int
2311 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2313 xfs_attr_leaf_name_local_t *name_loc;
2314 xfs_attr_leaf_name_remote_t *name_rmt;
2315 int size;
2317 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2318 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2319 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2320 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2321 be16_to_cpu(name_loc->valuelen));
2322 } else {
2323 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2324 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2326 return(size);
2330 * Calculate the number of bytes that would be required to store the new
2331 * attribute (whether local or remote only calculate bytes in this block).
2332 * This routine decides as a side effect whether the attribute will be
2333 * a "local" or a "remote" attribute.
2336 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2338 int size;
2340 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2341 if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2342 if (local) {
2343 *local = 1;
2345 } else {
2346 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2347 if (local) {
2348 *local = 0;
2351 return(size);
2355 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2358 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2360 attrlist_cursor_kern_t *cursor;
2361 xfs_attr_leafblock_t *leaf;
2362 xfs_attr_leaf_entry_t *entry;
2363 int retval, i;
2365 ASSERT(bp != NULL);
2366 leaf = bp->data;
2367 cursor = context->cursor;
2368 cursor->initted = 1;
2370 xfs_attr_trace_l_cl("blk start", context, leaf);
2373 * Re-find our place in the leaf block if this is a new syscall.
2375 if (context->resynch) {
2376 entry = &leaf->entries[0];
2377 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2378 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2379 if (cursor->offset == context->dupcnt) {
2380 context->dupcnt = 0;
2381 break;
2383 context->dupcnt++;
2384 } else if (be32_to_cpu(entry->hashval) >
2385 cursor->hashval) {
2386 context->dupcnt = 0;
2387 break;
2390 if (i == be16_to_cpu(leaf->hdr.count)) {
2391 xfs_attr_trace_l_c("not found", context);
2392 return(0);
2394 } else {
2395 entry = &leaf->entries[0];
2396 i = 0;
2398 context->resynch = 0;
2401 * We have found our place, start copying out the new attributes.
2403 retval = 0;
2404 for ( ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2405 attrnames_t *namesp;
2407 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2408 cursor->hashval = be32_to_cpu(entry->hashval);
2409 cursor->offset = 0;
2412 if (entry->flags & XFS_ATTR_INCOMPLETE)
2413 continue; /* skip incomplete entries */
2414 if (!xfs_attr_namesp_match_overrides(context->flags, entry->flags))
2415 continue;
2417 namesp = xfs_attr_flags_namesp(entry->flags);
2419 if (entry->flags & XFS_ATTR_LOCAL) {
2420 xfs_attr_leaf_name_local_t *name_loc =
2421 XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2423 retval = context->put_listent(context,
2424 namesp,
2425 (char *)name_loc->nameval,
2426 (int)name_loc->namelen,
2427 be16_to_cpu(name_loc->valuelen),
2428 (char *)&name_loc->nameval[name_loc->namelen]);
2429 if (retval)
2430 return retval;
2431 } else {
2432 xfs_attr_leaf_name_remote_t *name_rmt =
2433 XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2435 int valuelen = be32_to_cpu(name_rmt->valuelen);
2437 if (context->put_value) {
2438 xfs_da_args_t args;
2440 memset((char *)&args, 0, sizeof(args));
2441 args.dp = context->dp;
2442 args.whichfork = XFS_ATTR_FORK;
2443 args.valuelen = valuelen;
2444 args.value = kmem_alloc(valuelen, KM_SLEEP);
2445 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2446 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2447 retval = xfs_attr_rmtval_get(&args);
2448 if (retval)
2449 return retval;
2450 retval = context->put_listent(context,
2451 namesp,
2452 (char *)name_rmt->name,
2453 (int)name_rmt->namelen,
2454 valuelen,
2455 (char*)args.value);
2456 kmem_free(args.value, valuelen);
2458 else {
2459 retval = context->put_listent(context,
2460 namesp,
2461 (char *)name_rmt->name,
2462 (int)name_rmt->namelen,
2463 valuelen,
2464 NULL);
2466 if (retval)
2467 return retval;
2469 if (context->seen_enough)
2470 break;
2471 cursor->offset++;
2473 xfs_attr_trace_l_cl("blk end", context, leaf);
2474 return(retval);
2478 /*========================================================================
2479 * Manage the INCOMPLETE flag in a leaf entry
2480 *========================================================================*/
2483 * Clear the INCOMPLETE flag on an entry in a leaf block.
2486 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2488 xfs_attr_leafblock_t *leaf;
2489 xfs_attr_leaf_entry_t *entry;
2490 xfs_attr_leaf_name_remote_t *name_rmt;
2491 xfs_dabuf_t *bp;
2492 int error;
2493 #ifdef DEBUG
2494 xfs_attr_leaf_name_local_t *name_loc;
2495 int namelen;
2496 char *name;
2497 #endif /* DEBUG */
2500 * Set up the operation.
2502 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2503 XFS_ATTR_FORK);
2504 if (error) {
2505 return(error);
2507 ASSERT(bp != NULL);
2509 leaf = bp->data;
2510 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2511 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2512 ASSERT(args->index >= 0);
2513 entry = &leaf->entries[ args->index ];
2514 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2516 #ifdef DEBUG
2517 if (entry->flags & XFS_ATTR_LOCAL) {
2518 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2519 namelen = name_loc->namelen;
2520 name = (char *)name_loc->nameval;
2521 } else {
2522 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2523 namelen = name_rmt->namelen;
2524 name = (char *)name_rmt->name;
2526 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2527 ASSERT(namelen == args->namelen);
2528 ASSERT(memcmp(name, args->name, namelen) == 0);
2529 #endif /* DEBUG */
2531 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2532 xfs_da_log_buf(args->trans, bp,
2533 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2535 if (args->rmtblkno) {
2536 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2537 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2538 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2539 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2540 xfs_da_log_buf(args->trans, bp,
2541 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2543 xfs_da_buf_done(bp);
2546 * Commit the flag value change and start the next trans in series.
2548 error = xfs_attr_rolltrans(&args->trans, args->dp);
2550 return(error);
2554 * Set the INCOMPLETE flag on an entry in a leaf block.
2557 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2559 xfs_attr_leafblock_t *leaf;
2560 xfs_attr_leaf_entry_t *entry;
2561 xfs_attr_leaf_name_remote_t *name_rmt;
2562 xfs_dabuf_t *bp;
2563 int error;
2566 * Set up the operation.
2568 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2569 XFS_ATTR_FORK);
2570 if (error) {
2571 return(error);
2573 ASSERT(bp != NULL);
2575 leaf = bp->data;
2576 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2577 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2578 ASSERT(args->index >= 0);
2579 entry = &leaf->entries[ args->index ];
2581 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2582 entry->flags |= XFS_ATTR_INCOMPLETE;
2583 xfs_da_log_buf(args->trans, bp,
2584 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2585 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2586 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2587 name_rmt->valueblk = 0;
2588 name_rmt->valuelen = 0;
2589 xfs_da_log_buf(args->trans, bp,
2590 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2592 xfs_da_buf_done(bp);
2595 * Commit the flag value change and start the next trans in series.
2597 error = xfs_attr_rolltrans(&args->trans, args->dp);
2599 return(error);
2603 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2604 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2605 * entry given by args->blkno2/index2.
2607 * Note that they could be in different blocks, or in the same block.
2610 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2612 xfs_attr_leafblock_t *leaf1, *leaf2;
2613 xfs_attr_leaf_entry_t *entry1, *entry2;
2614 xfs_attr_leaf_name_remote_t *name_rmt;
2615 xfs_dabuf_t *bp1, *bp2;
2616 int error;
2617 #ifdef DEBUG
2618 xfs_attr_leaf_name_local_t *name_loc;
2619 int namelen1, namelen2;
2620 char *name1, *name2;
2621 #endif /* DEBUG */
2624 * Read the block containing the "old" attr
2626 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2627 XFS_ATTR_FORK);
2628 if (error) {
2629 return(error);
2631 ASSERT(bp1 != NULL);
2634 * Read the block containing the "new" attr, if it is different
2636 if (args->blkno2 != args->blkno) {
2637 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2638 -1, &bp2, XFS_ATTR_FORK);
2639 if (error) {
2640 return(error);
2642 ASSERT(bp2 != NULL);
2643 } else {
2644 bp2 = bp1;
2647 leaf1 = bp1->data;
2648 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2649 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2650 ASSERT(args->index >= 0);
2651 entry1 = &leaf1->entries[ args->index ];
2653 leaf2 = bp2->data;
2654 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2655 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2656 ASSERT(args->index2 >= 0);
2657 entry2 = &leaf2->entries[ args->index2 ];
2659 #ifdef DEBUG
2660 if (entry1->flags & XFS_ATTR_LOCAL) {
2661 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2662 namelen1 = name_loc->namelen;
2663 name1 = (char *)name_loc->nameval;
2664 } else {
2665 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2666 namelen1 = name_rmt->namelen;
2667 name1 = (char *)name_rmt->name;
2669 if (entry2->flags & XFS_ATTR_LOCAL) {
2670 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2671 namelen2 = name_loc->namelen;
2672 name2 = (char *)name_loc->nameval;
2673 } else {
2674 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2675 namelen2 = name_rmt->namelen;
2676 name2 = (char *)name_rmt->name;
2678 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2679 ASSERT(namelen1 == namelen2);
2680 ASSERT(memcmp(name1, name2, namelen1) == 0);
2681 #endif /* DEBUG */
2683 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2684 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2686 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2687 xfs_da_log_buf(args->trans, bp1,
2688 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2689 if (args->rmtblkno) {
2690 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2691 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2692 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2693 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2694 xfs_da_log_buf(args->trans, bp1,
2695 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2698 entry2->flags |= XFS_ATTR_INCOMPLETE;
2699 xfs_da_log_buf(args->trans, bp2,
2700 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2701 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2702 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2703 name_rmt->valueblk = 0;
2704 name_rmt->valuelen = 0;
2705 xfs_da_log_buf(args->trans, bp2,
2706 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2708 xfs_da_buf_done(bp1);
2709 if (bp1 != bp2)
2710 xfs_da_buf_done(bp2);
2713 * Commit the flag value change and start the next trans in series.
2715 error = xfs_attr_rolltrans(&args->trans, args->dp);
2717 return(error);
2720 /*========================================================================
2721 * Indiscriminately delete the entire attribute fork
2722 *========================================================================*/
2725 * Recurse (gasp!) through the attribute nodes until we find leaves.
2726 * We're doing a depth-first traversal in order to invalidate everything.
2729 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2731 xfs_da_blkinfo_t *info;
2732 xfs_daddr_t blkno;
2733 xfs_dabuf_t *bp;
2734 int error;
2737 * Read block 0 to see what we have to work with.
2738 * We only get here if we have extents, since we remove
2739 * the extents in reverse order the extent containing
2740 * block 0 must still be there.
2742 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2743 if (error)
2744 return(error);
2745 blkno = xfs_da_blkno(bp);
2748 * Invalidate the tree, even if the "tree" is only a single leaf block.
2749 * This is a depth-first traversal!
2751 info = bp->data;
2752 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2753 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2754 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2755 error = xfs_attr_leaf_inactive(trans, dp, bp);
2756 } else {
2757 error = XFS_ERROR(EIO);
2758 xfs_da_brelse(*trans, bp);
2760 if (error)
2761 return(error);
2764 * Invalidate the incore copy of the root block.
2766 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2767 if (error)
2768 return(error);
2769 xfs_da_binval(*trans, bp); /* remove from cache */
2771 * Commit the invalidate and start the next transaction.
2773 error = xfs_attr_rolltrans(trans, dp);
2775 return (error);
2779 * Recurse (gasp!) through the attribute nodes until we find leaves.
2780 * We're doing a depth-first traversal in order to invalidate everything.
2782 STATIC int
2783 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2784 int level)
2786 xfs_da_blkinfo_t *info;
2787 xfs_da_intnode_t *node;
2788 xfs_dablk_t child_fsb;
2789 xfs_daddr_t parent_blkno, child_blkno;
2790 int error, count, i;
2791 xfs_dabuf_t *child_bp;
2794 * Since this code is recursive (gasp!) we must protect ourselves.
2796 if (level > XFS_DA_NODE_MAXDEPTH) {
2797 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2798 return(XFS_ERROR(EIO));
2801 node = bp->data;
2802 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2803 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2804 count = be16_to_cpu(node->hdr.count);
2805 if (!count) {
2806 xfs_da_brelse(*trans, bp);
2807 return(0);
2809 child_fsb = be32_to_cpu(node->btree[0].before);
2810 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2813 * If this is the node level just above the leaves, simply loop
2814 * over the leaves removing all of them. If this is higher up
2815 * in the tree, recurse downward.
2817 for (i = 0; i < count; i++) {
2819 * Read the subsidiary block to see what we have to work with.
2820 * Don't do this in a transaction. This is a depth-first
2821 * traversal of the tree so we may deal with many blocks
2822 * before we come back to this one.
2824 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2825 XFS_ATTR_FORK);
2826 if (error)
2827 return(error);
2828 if (child_bp) {
2829 /* save for re-read later */
2830 child_blkno = xfs_da_blkno(child_bp);
2833 * Invalidate the subtree, however we have to.
2835 info = child_bp->data;
2836 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2837 error = xfs_attr_node_inactive(trans, dp,
2838 child_bp, level+1);
2839 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2840 error = xfs_attr_leaf_inactive(trans, dp,
2841 child_bp);
2842 } else {
2843 error = XFS_ERROR(EIO);
2844 xfs_da_brelse(*trans, child_bp);
2846 if (error)
2847 return(error);
2850 * Remove the subsidiary block from the cache
2851 * and from the log.
2853 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2854 &child_bp, XFS_ATTR_FORK);
2855 if (error)
2856 return(error);
2857 xfs_da_binval(*trans, child_bp);
2861 * If we're not done, re-read the parent to get the next
2862 * child block number.
2864 if ((i+1) < count) {
2865 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2866 &bp, XFS_ATTR_FORK);
2867 if (error)
2868 return(error);
2869 child_fsb = be32_to_cpu(node->btree[i+1].before);
2870 xfs_da_brelse(*trans, bp);
2873 * Atomically commit the whole invalidate stuff.
2875 if ((error = xfs_attr_rolltrans(trans, dp)))
2876 return (error);
2879 return(0);
2883 * Invalidate all of the "remote" value regions pointed to by a particular
2884 * leaf block.
2885 * Note that we must release the lock on the buffer so that we are not
2886 * caught holding something that the logging code wants to flush to disk.
2888 STATIC int
2889 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2891 xfs_attr_leafblock_t *leaf;
2892 xfs_attr_leaf_entry_t *entry;
2893 xfs_attr_leaf_name_remote_t *name_rmt;
2894 xfs_attr_inactive_list_t *list, *lp;
2895 int error, count, size, tmp, i;
2897 leaf = bp->data;
2898 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2901 * Count the number of "remote" value extents.
2903 count = 0;
2904 entry = &leaf->entries[0];
2905 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2906 if (be16_to_cpu(entry->nameidx) &&
2907 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2908 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2909 if (name_rmt->valueblk)
2910 count++;
2915 * If there are no "remote" values, we're done.
2917 if (count == 0) {
2918 xfs_da_brelse(*trans, bp);
2919 return(0);
2923 * Allocate storage for a list of all the "remote" value extents.
2925 size = count * sizeof(xfs_attr_inactive_list_t);
2926 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2929 * Identify each of the "remote" value extents.
2931 lp = list;
2932 entry = &leaf->entries[0];
2933 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2934 if (be16_to_cpu(entry->nameidx) &&
2935 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2936 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2937 if (name_rmt->valueblk) {
2938 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2939 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2940 be32_to_cpu(name_rmt->valuelen));
2941 lp++;
2945 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2948 * Invalidate each of the "remote" value extents.
2950 error = 0;
2951 for (lp = list, i = 0; i < count; i++, lp++) {
2952 tmp = xfs_attr_leaf_freextent(trans, dp,
2953 lp->valueblk, lp->valuelen);
2955 if (error == 0)
2956 error = tmp; /* save only the 1st errno */
2959 kmem_free((xfs_caddr_t)list, size);
2960 return(error);
2964 * Look at all the extents for this logical region,
2965 * invalidate any buffers that are incore/in transactions.
2967 STATIC int
2968 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2969 xfs_dablk_t blkno, int blkcnt)
2971 xfs_bmbt_irec_t map;
2972 xfs_dablk_t tblkno;
2973 int tblkcnt, dblkcnt, nmap, error;
2974 xfs_daddr_t dblkno;
2975 xfs_buf_t *bp;
2978 * Roll through the "value", invalidating the attribute value's
2979 * blocks.
2981 tblkno = blkno;
2982 tblkcnt = blkcnt;
2983 while (tblkcnt > 0) {
2985 * Try to remember where we decided to put the value.
2987 nmap = 1;
2988 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2989 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2990 NULL, 0, &map, &nmap, NULL, NULL);
2991 if (error) {
2992 return(error);
2994 ASSERT(nmap == 1);
2995 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2998 * If it's a hole, these are already unmapped
2999 * so there's nothing to invalidate.
3001 if (map.br_startblock != HOLESTARTBLOCK) {
3003 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3004 map.br_startblock);
3005 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3006 map.br_blockcount);
3007 bp = xfs_trans_get_buf(*trans,
3008 dp->i_mount->m_ddev_targp,
3009 dblkno, dblkcnt, XFS_BUF_LOCK);
3010 xfs_trans_binval(*trans, bp);
3012 * Roll to next transaction.
3014 if ((error = xfs_attr_rolltrans(trans, dp)))
3015 return (error);
3018 tblkno += map.br_blockcount;
3019 tblkcnt -= map.br_blockcount;
3022 return(0);
3027 * Roll from one trans in the sequence of PERMANENT transactions to the next.
3030 xfs_attr_rolltrans(xfs_trans_t **transp, xfs_inode_t *dp)
3032 xfs_trans_t *trans;
3033 unsigned int logres, count;
3034 int error;
3037 * Ensure that the inode is always logged.
3039 trans = *transp;
3040 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
3043 * Copy the critical parameters from one trans to the next.
3045 logres = trans->t_log_res;
3046 count = trans->t_log_count;
3047 *transp = xfs_trans_dup(trans);
3050 * Commit the current transaction.
3051 * If this commit failed, then it'd just unlock those items that
3052 * are not marked ihold. That also means that a filesystem shutdown
3053 * is in progress. The caller takes the responsibility to cancel
3054 * the duplicate transaction that gets returned.
3056 if ((error = xfs_trans_commit(trans, 0)))
3057 return (error);
3059 trans = *transp;
3062 * Reserve space in the log for th next transaction.
3063 * This also pushes items in the "AIL", the list of logged items,
3064 * out to disk if they are taking up space at the tail of the log
3065 * that we want to use. This requires that either nothing be locked
3066 * across this call, or that anything that is locked be logged in
3067 * the prior and the next transactions.
3069 error = xfs_trans_reserve(trans, 0, logres, 0,
3070 XFS_TRANS_PERM_LOG_RES, count);
3072 * Ensure that the inode is in the new transaction and locked.
3074 if (!error) {
3075 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
3076 xfs_trans_ihold(trans, dp);
3078 return (error);